A user's lengthy debugging resulted in them figuring out "Invite
emails" being disabled stopped the "/inv" command from sending invites
on discord, which makes sense except the confusing setting name (now
renamed "Messages" in the UI), and the fact that no error was reported.
This setting being disabled is now logged to the console when it's
attempted through the admin page or discord. For #378.
separated the search parts of ServerSearchReqDTO into
ServerFilterReqDTO, factored out part of GetActivities that made the
(filtering) query, which is then used in GetFilteredActivityCount with
db.Count.
all routes are registered on the root of the host, and the "reverse
proxy subfolder". When using Reverse proxy subfolder "/accounts", and
the default URL Path for the Admin page of "/" (a.k.a. ""), a route
collision occurs that for some reason wasn't dealt with before. This is
avoided by checking before registering the second route, and a panic
recovery handler has been added telling the user off for using bad
subpaths if any others occur.
URLs are made with app.ExternalURI (now included in window.pages), which
itself should include Base. Also fixed the subfolder being removed from
the url after login.
previously used app.jf.CacheExpiry = time.Now(), now either call
app.InvalidateJellyfinCache() (when we only need it to get a user),
or app.InvalidateUserCaches() (when the web user list needs to be
updated).
OffsetMinutesFromUTC was being passed incorrectly by the web app
(getTimezoneOffset if UTC - Timezone, we wanted Timezone - UTC), now
fixed. This value is now used if given in comparisons. Times are
truncated to minute-deep precision, and Any date comparison ignores
empty date fields (i.e. a unix timestamp being 0 or a time.Time.IsZero()
== true).
included UTC offset in minutes in DateAttempt, will be used shortly.
Also moved this stuff (ParsedDate, DateAttempt) to the common d.ts, and
the method for parsing from a string (now parseDateString) to common.
Also pre-emptively load the user cache when the admin page loads.
fixed activities sorting.
less oddities from me not knowing how to use flex boxes. Also kinda
standardised a button-inside-input thing, so added a .gap-<n> >
.inside-input selector which works on gap-1 and gap-2.
QueryDTO.Value being classed as "any" meant DateAttempts would be
unmarshaled as map[string]any, so a custom UnmarshalJSON checks the data
type and unmarshals into a DateAttempt if needed. mentionedUers,
referrer and time matching implemented for activity search. Also, fixed
multi-class queries (e.g. date -and- bool for last-active).
only supports queries of fields actually in Activity. The web UI only
directly queries two of these, ID and Time (mistakenly referneced as date in
the web ui previously). Later commits will come up with creative ways of
dealing with all the other query types.
forgot to wipe this._ordering, so search was attempting to run on
non-existent items. Also renamed the two functions to appendNewItems and
replaceWithNewItems, this makes more sense to me.
calculation for number of rows to be drawn was wrong, fixed now. To
compensate for overshooting with fast scrolling, speed is calculated
using previous scrollY in rows/scroll, and used to render more rows.
Also, the "scrollend" event is used to load more at the end of a scroll
always. Since this isn't available on safari/webkit(2gtk), a polyfill
has been added.
Found out the bottleneck when ~2000 or more elements are loaded isn't
the search or sort or anything, but the DOM. An infinite scroll
implementation is added, where elements are added to the DOM as you
scroll. May still be a little buggy, and can't yet cope with screen
resizes. Also, the "shown" indicator is broken.
Added ID/Name indexing to mediabrowser, and cleaned it up a little bit.
Was taking ~2s with 5000 users and firing tons of requests at Jellyfin,
now take ~160ms from cold boot, ~1ms with cache.
fixed bugs, added extra text on "no results found" to suggest server
searching, and conditionally disable the button based on search content
and current sort. Activities page still broken. Also fixed up cache
generation, only one should ever run now, as should sorting. Two time
thresholds exist, one to trigger a re-sync but do it in the background
(i.e. send the old one to the requester), and one to re-sync and wait
for it.
Pagination fully factored out, and both Activities and Accounts now use
a PaginatedList superclass. Server-side search is done by pressing enter
after typing a search, or by pressing the search button. Works on
accounts, soon will on activities if it doesn't already.
RecordCounter class created from that in activityList, and put in
accountsList. PageCount-type route standardized and made for /users
(/users/count). Created userCache, which regularly generates the
respUser list returned by /users. Added a currently dumb POST /users for
searching/pagination, GET /users is now just for getting -all- users.
go-getted expr, an expression language that seems like it'll be useful
for evaluating local searches. We don't store this data in the badger
DB, so we can't use the nice query form provided by badgerhold.
fixed activities sorting.
less oddities from me not knowing how to use flex boxes. Also kinda
standardised a button-inside-input thing, so added a .gap-<n> >
.inside-input selector which works on gap-1 and gap-2.
QueryDTO.Value being classed as "any" meant DateAttempts would be
unmarshaled as map[string]any, so a custom UnmarshalJSON checks the data
type and unmarshals into a DateAttempt if needed. mentionedUers,
referrer and time matching implemented for activity search. Also, fixed
multi-class queries (e.g. date -and- bool for last-active).
only supports queries of fields actually in Activity. The web UI only
directly queries two of these, ID and Time (mistakenly referneced as date in
the web ui previously). Later commits will come up with creative ways of
dealing with all the other query types.
forgot to wipe this._ordering, so search was attempting to run on
non-existent items. Also renamed the two functions to appendNewItems and
replaceWithNewItems, this makes more sense to me.
calculation for number of rows to be drawn was wrong, fixed now. To
compensate for overshooting with fast scrolling, speed is calculated
using previous scrollY in rows/scroll, and used to render more rows.
Also, the "scrollend" event is used to load more at the end of a scroll
always. Since this isn't available on safari/webkit(2gtk), a polyfill
has been added.
Harvey Tindall